International Journal of Epidemiology, 2014, 224–234
doi: 10.1093/ije/dyt259
Advance Access Publication Date: 14 January 2014
Cancer
Cancer
Is fluoride a risk factor for bone cancer? Small
area analysis of osteosarcoma and Ewing
sarcoma diagnosed among 0–49-year-olds in
Great Britain, 1980–2005
Karen Blakey,1 Richard G Feltbower,2 Roger C Parslow,2
Peter W James,1 Basilio Gómez Pozo,1,3 Charles Stiller,4 Tim J Vincent,4
Paul Norman,5 Patricia A McKinney,2 Michael F Murphy,4 Alan W Craft6
and Richard JQ McNally1*
1
Institute of Health and Society, Newcastle University, Newcastle upon Tyne, UK, 2Paediatric Epidemiology Group, University of Leeds, Leeds, UK, 3Clinical Management Unit of Preventive Medicine,
Surveillance and Health Promotion, Granada-Metropolitan Primary Health Care District, Andalusian
Health Service, Granada, Spain, 4Childhood Cancer Research Group, Department of Paediatrics,
University of Oxford, Oxford, UK, 5School of Geography, University of Leeds, Leeds, UK and 6Northern
Institute of Cancer Research, Newcastle University, Newcastle upon Tyne, UK
*Corresponding author. Institute of Health and Society, Newcastle University, Sir James Spence Institute, Royal Victoria
Infirmary, Queen Victoria Road, Newcastle upon Tyne NE1 4LP, UK. E-mail: [email protected]
Accepted 7 November 2013
Abstract
Background: Artificial fluoridation of drinking water to improve dental health has long
been a topic of controversy. Opponents of this public health measure have cited the possibility of bone cancer induction. The study objective was to examine whether increased
risk of primary bone cancer was associated with living in areas with higher concentrations of fluoride in drinking water.
Methods: Case data on osteosarcoma and Ewing sarcoma, diagnosed at ages 0–49 years
in Great Britain (GB) (defined here as England, Scotland and Wales) during the period
1980–2005, were obtained from population-based cancer registries. Data on fluoride levels in drinking water in England and Wales were accessed through regional water companies and the Drinking Water Inspectorate. Scottish Water provided data for Scotland.
Negative binomial regression was used to examine the relationship between incidence
rates and level of fluoride in drinking water at small area level.
Results: The study analysed 2566 osteosarcoma and 1650 Ewing sarcoma cases. There
was no evidence of an association between osteosarcoma risk and fluoride in drinking
water [relative risk (RR) per one part per million increase in the level of fluoride ¼ 1001;
90% confidence interval (CI) 0871, 1151] and similarly there was no association for
Ewing sarcoma (RR ¼ 0929; 90% CI 0773, 1115).
C The Author 2014. Published by Oxford University Press on behalf of the International Epidemiological Association
V
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0/),
which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited. For commercial re-use, please contact
224
[email protected]
225
International Journal of Epidemiology, 2014, Vol. 43, No. 1
Conclusions: The findings from this study provide no evidence that higher levels of fluoride (whether natural or artificial) in drinking water in GB lead to greater risk of either
osteosarcoma or Ewing sarcoma.
Key words: Osteosarcoma, Ewing sarcoma, bone cancer, children, young people, artificial fluoridation, fluoride,
drinking water, Great Britain, small area analysis
Key Messages
• There was no evidence of an association between fluoride in drinking water and osteosarcoma or Ewing sarcoma
before or after adjustment for small area level deprivation.
• 33% of artificially fluoridated water supply zones in Great Britain were found to be supplying water that was below
07 parts per million of fluoride, the lower limit of the optimal level for dental health benefit.
• There was no evidence that those who lived in an artificially fluoridated area of Great Britain were at increased risk
of osteosarcoma or Ewing sarcoma.
• There was no evidence that those living in an area of Great Britain with naturally occurring fluoride within the optimal
level for dental health benefit were at increased risk of osteosarcoma or Ewing sarcoma.
Introduction
Primary bone cancer is the third most common cancer in
10–24-year-olds, mainly comprising osteosarcoma and
Ewing sarcoma (ES).1 The aetiology is unclear, but both
genetic and environmental factors are likely to be
involved.1,2
Fluoride occurs naturally in drinking water at varying
concentrations. Water is a primary dietary fluoride source.
Populations supplied with high levels of naturally occurring fluoride in drinking water have low levels of dental
caries.3 The optimum range for dental health benefit is
07–12 parts per million (ppm). Concentrations below
03 ppm may provide no benefit.4
The recognition of an association between prevalence of
dental caries and areas of greater socio-economic deprivation led to artificial fluoridation of water supplies in some
countries, including in parts of Great Britain (GB), defined
here as England, Scotland and Wales.5 Pilot schemes established in the 1950s included: Kilmarnock (1956–62); part
of Anglesey (1955–92); Andover (1955–58) and Watford
(1956–89).6–8 Artificial fluoridation programmes were
introduced during the 1960s. Currently five water companies artificially fluoridate some drinking water supplies
under the instruction of health authorities (Figure 1).9
This practice has been controversial and there is speculation that adding fluoride to community water supplies
could result in adverse health outcomes including increased
risk of cancer, particularly osteosarcoma.10 Two systematic reviews concluded there was no clear association
between water fluoridation and osteosarcoma incidence or
mortality, but both advised caution in view of heterogeneity and other methodological concerns.11,12
The present study used the same case dataset as a previous demographic analysis that tested whether spatial variation among osteosarcoma and ES was associated with
population density and area-based deprivation. Area-based
deprivation is an ecological measure used for examining
socio-economic data according to a small geographical
unit. The previous ecological study found higher incidence
of osteosarcoma for females in less deprived areas and of
ES in areas of low population density and high car ownership. The putative association between osteosarcoma and
ES risk and fluoride in drinking water was not analysed.13
The present study objective was to examine whether
osteosarcoma risk was associated with fluoride levels in
drinking water. For comparison, ES was also studied. This
study is the first to analyse fluoride monitoring data as
part of the exposure measurement and the putative association with osteosarcoma risk in small geographical areas
across the whole of GB. No distinction was made between
artificial or naturally occurring fluoride in drinking water,
and novel geographical information system (GIS) methodologies were utilised to assign fluoride levels at small area
level. The following hypotheses were tested: (i) geographical heterogeneity of osteosarcoma incidence is modulated
by differences in fluoride levels in drinking water; and
(ii) geographical heterogeneity of ES incidence is not
modulated by differences in fluoride levels.
International Journal of Epidemiology, 2014, Vol. 43, No. 1
226
Key
Area of fluoride dosing
Boundary of artificially fluoridating
water company
1
All other areas
1
Northumbrian Water
2
United Utilities
3
Severn Trent Water
4
South Staffordshire Water
5
Anglian Water
2
3
4
5
© Crown Copyright and database right 2012.
Ordnance Survey Reference No. 1000022861
Figure 1. Map of England and Wales illustrating the boundaries of the water companies with artificial fluoridation programmes in place.
Methods
Study subjects
Data for patients diagnosed with osteosarcoma or ES in
GB during 1980–2005 were analysed. Cases were limited
to ages 0–49 years, as there are few ES cases above these
ages and osteosarcoma over the age of 50 years is usually
associated with Paget’s disease or is secondary to radiotherapy.2,14 To ensure sufficient case numbers in each category at small area level, cases were sub-divided into age
groups 0–14, 15–29 and 30–49 years at diagnosis.
Regulatory and ethical approvals were gained (UK
National Research Ethics Service reference number
09/H0904/5). Case data were accessed from the 10 regional cancer registries in GB. Case data from the National
Registry of Childhood Tumours were extracted and used
to cross-check accuracy of the regional registry cases aged
0–14 years.
Diagnostic groups
Cases were grouped using the International Classification
of Diseases for Oncology, third edition (ICD-O-3).15
Osteosarcoma and ES were specified a priori and the associated topography and morphology codes are given
elsewhere.13
227
Population data
Denominator data were derived from national decennial
census data.16–21 Analyses over a prolonged time span are
impeded by boundary changes, especially at small area
level.22 Therefore, population counts from previous
censuses and geo-referenced bone cancer registration data
were adjusted to be compatible with 2001 census boundaries.23 The small area units (SAU) used in the analyses were
census wards in England and Wales (0–49-years-old population ranges from 297 to 29 300, median ¼ 3090), and
postcode sectors in Scotland (0–49-years-old population
ranges from 23 to 15 916, median ¼ 3201).
Similarly, to adjust for deprivation, a time-series of indicators was obtained from each census during the study
period and geographically converted to be compatible with
2001 SAUs. The Townsend index (comprising four components: unemployment, non-car ownership, non-home ownership and household overcrowding) is commonly used in
health studies.24 To track changes in deprivation for every
SAU at the different time points, each variable was
expressed as a z-score relative to the GB average level over
the study period, then summed and equally weighted to a
single deprivation score.25
Fluoride monitoring data
At the time of the study, 25 companies in GB supplied
water to 2265 statutorily demarcated ‘water supply zones’
(WSZs) with a population supply threshold of 100 000
(Water Regulation Zones in Scotland). Less than 1% of
households have private water supplies.26 Fluoride level in
drinking water is continuously monitored and required to
be less than 15 ppm on a 3–month average basis.27,28
These routine fluoride monitoring data were utilised in this
study and obtained through Scottish Water and a regulatory body, the Drinking Water Inspectorate (DWI) for
companies in England and Wales.
Digital boundary data
Digital boundaries for census areas facilitated georeferenced data linkage for statistical analysis and were
accessed from UK Borders.29,30 Digital boundaries for
WSZs were accessed through the DWI and Scottish Water.
Data linkage: assignment of fluoride level to
census small area units
Postcode distributions were used to link WSZs to each
SAU which was subsequently assigned a fluoride level. The
population centroid of each SAU was calculated in ESRI
International Journal of Epidemiology, 2014, Vol. 43, No. 1
ArcMap 9.331 to assign a weighted average for SAUs with
water supply from more than one WSZ. The Nondetects
and Data Analysis (NADA) add-on package for estimating
censored environmental data was used to compute values
for left-censored or missing data in R 281.32,33 Bone cancer cases were linked and aggregated for each 2001 census
SAU.
Statistical analysis
Negative binomial regression was implemented using
STATA version 12.34 The logarithm of the incidence rate
was modelled in two parts: (i) variation explained by
the data was expressed as a linear function of gender,
age-group and deprivation; (ii) unexplained variation was
modelled by a negative binomial distribution. The number
of cases observed in each SAU was the dependent variable
and the logarithm of the ‘at risk’ population was used as
the offset. Independent variables were the census-derived
SAU attributes that were allocated to the 2001 census
geography.22
The previously determined best-fitting models that explained bone cancer variation (the final models in the hierarchical series for osteosarcoma and ES)13 were used as the
base models. In the present study, these models were extended to include fluoride level in drinking water by SAU.
For osteosarcoma, the base model included gender,
age-group, the interaction gender*age-group, the Townsend score and the interaction Townsend*female. For ES,
the base model included age-group, gender, the interaction
age-group*gender, Scotland, East Midlands, population
density and non-car ownership. Fluoride, which had not
been analysed in the previous analysis, was then added to
these base models. The effect of fluoride level in drinking
water was tested in multivariable models whilst adjusting
for gender, age-group, population density and interactions.
The effect of fluoride was assessed, after adjustment for
these covariates, using likelihood ratio tests and the Akaike
Information Criterion (AIC).35 Relative risks (RRs) and
associated 90% confidence intervals (CIs) were
calculated.36
The mean fluoride level in drinking water was determined using monitoring data sampled between 2004 and
2006 and modelled as a continuous variable, under the assumption that any association was linear. The effect of
fluoride level in drinking water was modelled with and
without adjustment for deprivation. To test the linearity
assumption, fluoride was also modelled as an ordinal variable (dividing the distribution into fifths). To test for a possible threshold effect, the most fluoridated fifth of the
population was compared with less fluoridated fifths.
Also, areas with a level of at least 07 ppm, the lower limit
International Journal of Epidemiology, 2014, Vol. 43, No. 1
of the range for optimum dental health benefit4 were compared with less fluoridated areas.
Sensitivity analyses tested for effects of age-group. The
data were restricted to cases born within the following
non-overlapping cohorts: (i) before 1970; (ii) 1970–79;
and (iii) 1980 onwards. To test for constancy of fluoride
exposures in regions where levels may have fluctuated,
artefactually inflated levels (1 ppm) were analysed. Finally,
an analysis was carried out on the osteosarcoma data using
age-groups 0–9, 10-14,. . .45-49 years to assess whether the
three age-groups 0–14, 15–29 and 30–49 years were too
broad to detect increases limited to specific ages. This additional analysis examined age-group interactions with
fluoride to test if the age-group 0–9 years was more
affected by fluoride than other age-groups. For all analyses, P-values were two-sided.
Results
The study analysed 2566 osteosarcoma cases (1493 males,
1073 females) and 1650 ES cases (988 males, 662 females)
aged 0–49 years. The numbers of cases by age-group and
gender are given in Table 1, with full descriptive data
given elsewhere.13 For osteosarcoma the overall agestandardized rate (ASR) for all persons was 264 per million persons per year (90% CI 255, 272), and 176 per
million persons per year (90% CI 168, 183) for ES. Most
analyses included an adjustment for deprivation.
After adjustment for gender, age-group, the interaction
gender*age-group, the Townsend score and the interaction
Townsend*female, no association was found between
osteosarcoma and fluoride levels in drinking water
(P ¼ 0987). The RR for osteosarcoma for 1-ppm increase
in fluoride level was 1001 (90% CI 0871 to 1151). After
adjustment for age-group, gender, the interaction
age-group*gender, Scotland, East Midlands, population
density and non-car ownership, no association was found
between ES incidence and fluoride levels in drinking water
(P ¼ 0503). The RR for ES for 1-ppm increase in fluoride
levels was 0929 (90% CI 0773 to 1115).
There was no effect without adjustment for deprivation.
Similarly, using ordinal measures there was no evidence of
any threshold effect. When testing cohorts born before
1970, 1970–79 and 1980 onwards, there was no differential effect of cohort on the association between bone cancer
and fluoride. An effect was not found when testing with
artefactually raised levels of fluoride (Tables 2–4), nor
when using narrower age-bands (0–9, 10–14. . .45–49
years). There was no evidence of an interaction between
age-group and fluoride.
Table 5 presents the mean fluoride level assigned to
each SAU in GB. The means range from 000 to 126 ppm;
228
Table 1. Number of cases of osteosarcoma and Ewing sarcoma by age-group and gender
Age-group
(years)
0-14
15-29
30-49
0-49
Number of
osteosarcoma
cases
Number of
Ewing sarcoma
cases
Males
Females
Total
Males
Females
Total
406
821
266
1493
411
494
168
1073
817
1315
434
2566
356
516
116
988
303
284
75
662
659
800
191
1650
48% of the SAUs had a fluoride level less than 010 ppm
and 80% had a level below 026 ppm. Approximately 70%
of the SAUs belonging to the upper fifth had fluoride levels
considered to provide the optimal dental decay prevention
benefit (070–120 ppm). This equated to approximately
14% of the total number of SAUs in GB having a fluoride
level within optimal limits. Five water companies in England currently operate artificial fluoridation programmes
but not all their WSZs are artificially fluoridated (AF)
(Figure 1). The AF WSZs ranged from 004 to 108 ppm;
67% of the AF WSZs had a level within optimal limits.
Discussion
This ecological analysis used high-quality populationbased osteosarcoma and ES case data from 0–49-year-olds
diagnosed 1980–2005 in GB. The demographic profile of
the study population has previously been published.13
There was no evidence of an association between fluoride
in drinking water and osteosarcoma or ES. Thus, there was
no support for prior hypothesis (i) that geographical heterogeneity of osteosarcoma is modulated by differences in
fluoride levels. There was support for prior hypothesis (ii)
that geographical heterogeneity of ES is not modulated by
differences in fluoride levels.
This is the first study that has assessed fluoride levels in
drinking water across the whole of GB. Novel methodologies were developed within a GIS framework to enable
fluoride levels to be assigned to each SAU. Such an approach reduced the potential of misclassification of exposure data when compared with previous studies that took
simpler approaches.37–39
The monitoring data suggested that levels in some AF
areas were much lower than 1 ppm. Indeed, 33% of AF
WSZs were below 07 ppm, the lower limit of the optimum
range for dental health benefit.4 It is noteworthy that this
corresponded to only 14% of all SAUs in GB having a
fluoride level that may confer a dental health benefit, and
229
International Journal of Epidemiology, 2014, Vol. 43, No. 1
Table 2. Description of models used in the sensitivity analyses that tested the assumption that: (a) a continuous scale is an
appropriate level of measurement for quantity of fluoride in drinking water; (b) there was no change in the effect of fluoride by
cohort; and (c) fluoride levels in drinking water are constant over the study period
Model
Test
Factora,b
(a)
Fluoride level in drinking water as a continuous variable with adjustment for deprivation
Fluoride level in drinking water as a continuous variable without adjustment for deprivation
Fluoride level in drinking water as a discrete variable (quintiles) with adjustment for deprivation and comparison of
quintile 1 with quintile 2, 3, 4 and 5
Fluoride level in drinking water as a binary variable based on the highest (5th) quintile with adjustment for deprivation
Fluoride level in drinking water higher than 07 ppm are compared to levels that are less than 07 ppm (where fluoride level
as binary based on 07 ppm threshold) with adjustment for deprivation
Fluoride level in drinking water as a discrete variable (quintiles) without adjustment for deprivation and comparison of
quintile 1 with quintile 2, 3, 4 and 5
Fluoride level in drinking water as a binary variable based on the highest (5th) quintile without adjustment for deprivation
Fluoride levels in drinking water higher than 07 ppm are compared with levels that are less than 07 ppm (where fluoride
level as binary is based on 07 ppm threshold) without adjustment for deprivation
(b)
Fluoride level in drinking water (where cohort is restricted to include cases born before 1970)
Fluoride level in drinking water (where cohort is restricted to include cases born 1970 to 1979)
Fluoride level in drinking water (where cohort is restricted to include cases born 1980 or later)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
(c)
19
20
21
22
23
Fluoride level in drinking water (where 10% of AF SAUs assigned a fluoride level of 1 ppm)
Fluoride level in drinking water (where 20% of AF SAUs assigned a fluoride level of 1 ppm)
Fluoride level in drinking water (where 30% of AF SAUs assigned a fluoride level of 1 ppm)
Fluoride level in drinking water (where 40% of AF SAUs assigned a fluoride level of 1 ppm)
Fluoride level in drinking water (where 100% of AF SAUs assigned a fluoride level of 1 ppm)
Fluoride level in drinking water (where pilot areas Watford, Kilmarnock & all Anglesey assigned a fluoride level of 1 ppm)
Fluoride level in drinking water (where pilot areas Watford, Kilmarnock & wards in north Anglesey assigned a fluoride
level of 1 ppm)
Fluoride level in drinking water (where pilot areas Watford, Kilmarnock & wards in south Anglesey assigned a fluoride
level of 1 ppm)
Fluoride level in drinking water (where pilot areas Watford, Kilmarnock & wards in east Anglesey assigned a fluoride level
of 1 ppm)
Fluoride level in drinking water (where pilot areas Watford, Kilmarnock & wards in west Anglesey assigned a fluoride level
of 1 ppm)
Fluoride level in drinking water (excluding all AF SAUs including those in Three Valleys)
Artificial fluoridation as a binary variable with adjustment for deprivation
a
The best fitting model for osteosarcoma is age-group, gender, gender*age-group, Townsend & Townsend*female.13
The best fitting model for Ewing sarcoma is age-group, gender, age-group*gender, Scotland, East Midlands, population density and non-car ownership.13
b
61% of AF SAUs had such a level. This suggests that 35%
of populations residing in AF areas were being supplied
with AF water dosed below the optimal level.
The relationship between fluoride and osteosarcoma
risk has been examined in a small number of animal40,41
and human studies37–39,41–48 with conflicting results. Disagreement could be linked to fundamental differences in
study design; some studies were laboratory-based,40,41
some ecological,37,38,43–45 others were case-control.46–48
Moreover, investigations to date had several methodological limitations. These included limited statistical
power,37,46–48 the potential of selection bias in choice of
cases47,48 and controls48 or method of exposure categorization leading to misclassification.
In a recent case-control study, Bassin and colleagues
analysed 103 cases (60 males, 43 females) aged under
20 years and found increased osteosarcoma risk with
fluoride in drinking water for males only, with a peak in
the age-group 6–8 years.47 However, the number of cases
within this age-group would have been extremely
small.49,50 A further limitation acknowledged by the
authors was potential selection bias because of differences
in case and control referral patterns to participating
hospitals.47
International Journal of Epidemiology, 2014, Vol. 43, No. 1
230
Table 3. Examination of the effect of fluoride level in drinking water on osteosarcoma incidence: comparison of fluoride models
and test results
dfb
Model
deviance AICc
Numbera Comparison
d
1
2
e
f
Difference in:
df
175665 179409 247548 1
175667 179763 247862 1
0000
0006
3
175662 179365 247564 4
4368
4
5
d
175665 179406 247545 1
175665 179408 247547 1
0273
0114
6f
e
175664 179733 247891 4
3051
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
e
175667
175667
175667
175665
117108
175665
175665
175665
175665
175665
175665
175665
175665
175665
175665
148137
175667
175665
117108
0440
0367
3425
0965
1656
0024
0066
0010
0021
0030
0006
0017
0005
0002
0000
1090
3425
0160
3216
e
g
h
i
d
d
d
d
d
d
d
d
d
d
j
g
23
h
i
179759
179759
86698
65873
64391
179409
179408
179409
179409
179408
179409
179409
179409
179409
179409
149049
86698
65881
64375
247858
247858
114627
86060
86039
247547
247547
247548
247547
247547
247548
247547
247548
247548
247548
205272
114008
86068
86039
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
Relative risk (with 90%
confidence interval)
deviance
d
d
Coefficient (with 90%
confidence interval)
(a)
(b)
(c)
(d)
(a)
(b)
(c)
(d)
0001 (0138,0141)
0006 (0145,0133)
0122 (0227,0016)
0097 (0202,0009)
0056 (0160,0048)
0042 (0146,0061)
0026 (0056,0108)
0024 (0140,0093)
0102 (0207,0004)
0038 (0142,0067)
0058 (0161,0046)
0016 (0119,0088)
0033 (0049,0115)
0043 (0159,0074)
0193 (0421,0036)
0148 (0097,0393)
0197 (0051,0445)
0013 (0150,0124)
0021 (0158,0115)
0008 (0141,0125)
0011 (0118,0141)
0013 (0106,0131)
0006 (0131,0143)
0011 (0126,0148)
0006 (0143,0132)
0004 (0134,0141)
0001 (0136,0139)
0156 (0086,0397)
0189 (0361,0017)
0045 (0139,0229)
0197 (0020,0373)
(a)
(b)
(c)
(d)
(a)
(b)
(c)
(d)
1.001 (0.871,1.151)
0994 (0865,1142)
0885 (0797,0984)
0908 (0817,1009)
0946 (0853,1049)
0958 (0864,1063)
1026 (0946,1114)
0976 (0869,1097)
0903 (0813,1004)
0963 (0867,1069)
0944 (0851,1047)
0984 (0888,1092)
1034 (0952,1122)
0958 (0853,1076)
0825 (0656,1036)
1160 (0908,1482)
1218 (0951,1560)
0987 (0860,1132)
0979 (0854,1122)
0992 (0868,1133)
1011 (0889,1151)
1013 (0899,1140)
1006 (0877,1154)
1011 (0881,1160)
0994 (0866,1141)
1004 (0875,1152)
1001 (0873,1149)
1168 (0917,1488)
0828 (0697,0983)
1046 (0870,1257)
1218 (1021,1452)
Likelihood
ratio:
P-value
0987
0940
0359
0602
0736
0549
0507
0545
0064
0326
0198
0876
0798
0921
0885
0862
0941
0896
0947
0963
0986
0297
0064
0689
0073
a
Please see Table 2a–c for model description.
Degree of freedom.
c
Akaike Information Criterion
d
Compared with the non-fluoride model that contained age-group, gender, gender*age-group, Townsend & Townsend*female—cohort includes all cases in
GB for the whole of the study period.
e
Compared with the non-fluoride model that contained age-group, gender, gender*age-group—cohort includes all cases in GB for the whole of the study
period.
f
Coefficients and relative risks are reported for quintiles (a) 2, (b) 3, (c) 4 and (d) 5.
g
Compared with the non-fluoride model that contained age-group, gender, unemployment—cohort is restricted to include cases born before 1970.
h
Compared with the non-fluoride model that contained age-group, gender, gender*age-group, non-home ownership, non-home ownership *age-group—cohort
is restricted to include cases born between 1970 and 1979.
i
Compared with the non-fluoride model that contained age-group, gender, gender*age-group, unemployment, unemployment*age-group—cohort is restricted
to include cases born 1980 or later.
b
Another case-control study using the same dataset
found no association, but again methods used introduced
limitations. There were 109 cases aged under 30 years, but
only 21 controls in this age range (median age for male
controls ¼ 413 years; male cases ¼ 170).48 It is also
possible that the use of other newly diagnosed malignant
bone tumours as the controls masked any differences if risk
also increased in those tumours.21 Furthermore, although
the overall results contradict those from Bassin’s study,51
the use of total accumulated fluoride dose rather than a
231
International Journal of Epidemiology, 2014, Vol. 43, No. 1
Table 4. Examination of the effect of fluoride level in drinking water on Ewing sarcoma incidence: comparison of fluoride models and test results
dfb
Model
deviance AICh
Numbera Comparison
d
1
2
e
f
Difference in:
df
175663 128748 173016
175664 128830 173078
1
1
0448
0617
3
175660 128696 173024
4
5630
4
5
d
175663 128752 173020
175663 128749 173017
1
1
0039
0351
6f
e
175661 128802 173110
4
3416
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
e
175664
175664
175666
175665
117108
175663
175663
175663
175663
175663
175663
175663
175663
175663
175663
148135
175666
175665
117108
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
0002
0761
0679
0164
<0001
0564
0586
0807
0640
0780
0396
0513
0420
0520
0413
0037
1627
0010
<0001
e
g
h
i
d
d
d
d
d
d
d
d
d
d
j
g
23
h
i
128836
128828
51381
52883
51270
128747
128746
128747
128747
128745
128748
128747
128748
128747
128748
107981
51371
52885
51270
173084
173076
65437
67974
67076
173015
173014
173012
173014
173012
173016
173015
173016
173015
173016
144749
65427
67956
67076
Relative risk (with 90%
confidence interval)
deviance
d
d
Coefficient (with 90%
confidence interval)
(a)
(b)
(c)
(d)
(a)
(b)
(c)
(d)
0074 (0257,0109)
0086 (0269,0096)
0071 (0068,0211)
0120 (0268,0027)
0019 (0174,0136)
0002 (0150,0146)
0013 (0094,0120)
0053 (0202,0095)
0069 (0071,0209)
0076 (0222,0070)
0028 (0126,0182)
0011 (0137,0159)
0003 (0104,0110)
0078 (0225,0070)
0152 (0460,0156)
0072 (0366,0222)
0022 (0293,0337)
0082 (0262,0098)
0083 (0261,0096)
0095 (0271,0080)
0083 (0253,0088)
0083 (0240,0073)
0068 (0248,0111)
0078 (0259,0103)
0071 (0251,0110)
0079 (0260,0102)
0070 (0251,0110)
0039 (0294,0372)
0178 (0412,0057)
0013 (0228,0202)
0001 (0229,0227)
(a)
(b)
(c)
(d)
(a)
(b)
(c)
(d)
0929 (0773,1115)
0917 (0764,1101)
1074 (0934,1235)
0887 (0765,1028)
0981 (0840,1146)
0998 (0861,1157)
1013 (0910,1128)
0948 (0817,1100)
1071 (0931,1233)
0927 (0801,1073)
1028 (0881,1199)
1011 (0872,1173)
1003 (0901,1116)
0925 (0798,1073)
0859 (0631,1168)
0930 (0693,1249)
1023 (0746,1401)
0921 (0769,1103)
0921 (0770,1101)
0909 (0763,1084)
0921 (0776,1092)
0920 (0787,1076)
0934 (0780,1118)
0925 (0772,1108)
0932 (0778,1116)
0924 (0771,1108)
0932 (0778,1117)
1040 (0745,1451)
0837 (0662,1058)
0987 (0796,1223)
0999 (0796,1255)
Likelihood
ratio:
P-value
0503
0432
0229
0843
0554
0491
0967
0383
0410
0685
0996
0453
0444
0369
0424
0377
0529
0474
0517
0471
0520
0847
0202
0919
0996
a
Please see Table 2a–c for model description.
Degree of freedom.
c
Akaike Information Criterion.
d
Compared with the model that contains age-group, gender, age-group *gender, Scotland, East Midlands, population density and non-car ownership—cohort
includes all cases for whole of study period.
e
Compared with the model that contains age-group, gender, age-group *gender, Scotland, East Midlands, population density—cohort includes all cases for
whole of study period.
f
Coefficients and relative risks are reported for quintiles (a) 2, (b) 3, (c) 4 and (d) 5.
g
Compared with the non-fluoride model that contained age-group, gender, population density, unemployment—cohort is restricted to include cases born before
1970.
h
Compared with the non-fluoride model that contained age-group, gender, age-group *gender, population density, unemployment—cohort is restricted to include cases born between 1970 and 1979.
i
Compared with the non-fluoride model that contained age-group, gender, age-group *gender, Scotland, unemployment—cohort is restricted to include cases
born 1980 or later.
b
specific time in the life course prevents any direct comparisons being made.
A study did find a link between mean fluoride levels in
blood serum and increased osteosarcoma risk, but finding
this association does not infer causality.52 Mean fluoride
levels in cases with other bone-forming tumours were significantly higher when compared with the control group
consisting of patients with musculo-skeletal pain only.
International Journal of Epidemiology, 2014, Vol. 43, No. 1
Table 5. The upper and lower limits of each fluoride category
(the average level of fluoride assigned to each census small
area unit (SAU) in parts per million (ppm) and then the SAU
population distribution ranked and divided into fifths)
Fluoride category
(SAU population distribution
divided into fifths)
Average level of fluoride assigned
to each census SAU (ppm)
(upper and lower limits of
fluoride category)
1 (lower fifth)
2
3
4
5 (upper fifth)
0000000 – 0048969
0048970 – 0078770
0078771 – 0138820
0138821 – 0254040
0254041 – 1268000
Although the mean fluoride level was only approximately
50% of the osteosarcoma group, it highlights caution
when selecting controls.
Two ecological studies from the USA and from Ireland
concluded water fluoridation status had no influence on
osteosarcoma incidence rates.38,39 However, both studies
had shortcomings due to the methods of exposure data categorization. The US study categorized states according to
high and low community water fluoridation.37 Other studies merely compared bone cancer incidence in areas with
and without artificial fluoridation programmes.38,39 The
bioavailability and chemistry have been assessed, with no
difference being found between naturally occurring and
artificial fluoride.53 Similarly, another experimental double-blind cross-over trial reported no difference in bioavailability although this finding needed to be treated
cautiously due to small case numbers.54
The present study had some limitations. It is a small
area study and assumed that the characteristics of each individual within a designated area were represented by the
aggregate statistics for that area. However, although these
studies are susceptible to the ecological fallacy,55 ecological analyses are suited to initial investigation of causal
hypotheses.56 Artificial fluoridation programmes are necessarily ecological but are considered an effective method
of providing populations with the dental health benefit of
fluoride.22,23
Lack of availability and inconsistency of individual
sampling data across the whole of GB during the study
period (only 2004–06 data were used) meant an assumption was made of no change in fluoride levels within the
study time-frame. However, this assumption was substantiated through sensitivity analyses (Tables 2-4). The fluctuations in fluoride levels over time were assessed. Fluoride
levels in England fluctuated plus or minus 10%. Since
source data for Scotland were limited to 2004–06 and
232
more variation was found in Wales, the main analysis was
repeated using data for England only. It found the bestfitting models had the same predictors as the whole of GB.
It is well established that fluoride concentrations in
bone increase with age.57 To determine whether age or
time of putative exposure had influenced the findings, the
analyses were repeated using non-overlapping birth cohorts (before 1970, 1970–79 and 1980 onwards). For all
three cohorts, there was no evidence of any association
with risk (Tables 2–4). This confirmed that making an assumption of stable fluoride concentrations over time was
reasonable.
Finding no association might be because of attenuation
due to exposure measurement error, arising through the
imprecision in allocating fluoride levels to specific areas
during specific periods. For example, the WSZ boundaries
have changed over time but it was not possible to represent
these changes, as digital boundary data have only been
archived since 2004. Lack of data availability also made it
impossible to take any local changes within artificial fluoridation supply areas into account.
SAU of residence at time of diagnosis may not represent
the true lifetime or shorter period of fluoride exposure for
each case. Other sources of fluoride are not taken into consideration. In the 1950s when the first pilot studies were
carried out in GB, there was much less availability from
other sources. Fluoride started to be added to toothpaste in
the 1970s and by 1978 approximately 96% of all toothpastes were fluoridated. Nevertheless, it is believed that
drinking water is still the primary source of fluoride in GB,
particularly in areas with fluoride levels of 1 ppm and
over.4
In conclusion, this small area analysis used high-quality
population-based osteosarcoma and ES case data. Novel
GIS methodologies were developed to enable fluoride level
in drinking water to be assigned to each SAU in GB. No association was found between fluoride level and osteosarcoma or ES before and after adjustment for deprivation.
The findings from this study provide no evidence that
higher levels of fluoride (whether natural or artificial) in
drinking water in GB lead to greater risk of osteosarcoma
or ES. Ecological design was appropriate for this initial investigation but also introduced limitations. Further research, such as large case-control studies that incorporate
the GIS methodologies developed during this study, is
recommended.
Funding
This work was supported by the Bone Cancer Research Trust
(BCRT) [grant reference number: BCRT/08/08], North of England
Children’s Cancer Research (NECCR) Fund, the National Institute
for Health Research (NIHR) and Children with Cancer UK.
233
Acknowledgements
We are most grateful to Public Health England Knowledge and
Intelligence Team (West Midlands, formerly known as West Midlands Cancer Intelligence Unit), in particular to Dr Gill Lawrence
and Sally Vernon who co-ordinated the data request from cancer
registries in England and Wales. Likewise, we would like to thank
the Scottish Registry for facilitating the acquisition of case data for
Scotland. We are indebted to David Martin, Professor of Geography
at the University of Southampton, for his guidance with geo-conversion methodologies and also to Richard Hardy (funded by the
NECCR at the Institute of Health and Society, Newcastle University) for providing information technology support. Finally, we
would also like thank all water companies in GB who provided data
and the DWI, particularly Andy Taylor, Water Quality Data Manager and Mark Kozlowski, for construction of the map. The work is
based on census data which are copyright of the Crown. The work
is also based on data provided with the support of the ESRC and
Joint Information Systems Committee (JISC) and uses census boundary material, which is copyright of the Crown and the ED-LINE
Consortium.
Conflict of interest: None declared.
References
1. Eyre R, Feltbower RG, Mubwandarikwa E, Eden TO, McNally
RJ. Epidemiology of bone tumours in children and young adults.
Pediatr Blood Cancer 2009;53:941–52.
2. Savage SA, Mirabello L. Using epidemiology and genomics
to understand osteosarcoma etiology. Sarcoma 2011;
2011:548151.
3. Ainsworth N. Mottled teeth. Br Dent J 1933;55:233–50.
4. Harrison PT. Fluoride in water: A UK perspective. J Fluor Chem
2005;126:1448–56.
5. Pizzo G, Piscopo MR, Pizzo I, Giuliana G. Community water
fluoridation and caries prevention: a critical review. Clin Oral
Invest 2007;11:189–93.
6. Ministry of Health, Scottish Office, Ministry of Housing and
Local Government. The Conduct of the Fluoridation Studies in
the United Kingdom and the Results Achieved After Five Years.
London: HMSO, 1962.
7. Department of Health and Social Security, Scottish Office,
Ministry of Housing and Local Government. The Fluoridation
Studies in the United Kingdom and the Results Achieved After
Eleven Years. London: HMSO, 1969.
8. Murray JJ, Nunn JH, Steele JG (eds). Prevention of Oral
Disease. New York: Oxford University Press, 2003.
9. Mullen J, European Association for Paediatric Dentistry. History
of water fluoridation. Br Dent J 1995;199(Suppl 7):1–4.
10. Committee to Coordinate Environmental Health and Related
Programs, Ad Hoc Subcommittee on Fluoride. Review of
Fluoride: Benefits and Risks. Washington, DC: Public Health
Service, Department of Health and Human Services, 1991.
11. McDonagh MS, Whiting PF, Wilson PM et al. Systematic review
of water fluoridation. BMJ 2000;321:855–59.
12. Yeung CA. A systematic review of the efficacy and safety of
fluoridation. Evid Based Dent 2008;9:39.
13. McNally RJ, Blakey K, Parslow RC et al. Small area analyses of
bone cancer diagnosed in 0-49 year olds from Great Britain provide clues to aetiology. BMC Cancer 2012;12:270.
International Journal of Epidemiology, 2014, Vol. 43, No. 1
14. Ottaviani G, Jaffe N. The epidemiology of osteosarcoma. Canc
Treat Res 2009;152:3–13.
15. Fritz A, Percy C, Jack A et al. International Classification of
Diseases for Oncology (ICD-O). 3rd edn. Geneva: World Health
Organization, 2000.
16. Office of Population Censuses and Surveys. 1981 Census: Small
Area Statistics (England and Wales). Manchester, UK: Census
Dissemination Unit, Mimas, University of Manchester, 1981.
17. Registrar General for Scotland. 1981 Census: Small Area
Statistics (Scotland). Manchester, UK: Census Dissemination
Unit, Mimas, University of Manchester, 1981.
18. Office for Population Censuses and Surveys. 1991 Census: Small
Area Statistics (England and Wales). Manchester, UK: Census
Dissemination Unit, Mimas, University of Manchester, 1991.
19. General Register Office Scotland. 1991 Census: Small Area
Statistics (Scotland). Manchester, UK: Census Dissemination
Unit, Mimas, University of Manchester, 1991.
20. Office for National Statistics. 2001 Census: Standard Area
Statistics (England and Wales). Manchester, UK: Census
Dissemination Unit, Mimas, University of Manchester, 2001.
21. General Register Office for Scotland. 2001 Census: Standard
Area Statistics (Scotland). Manchester, UK: Census Dissemination Unit, Mimas, University of Manchester, 2001.
22. Simpson L. Geography conversion tables: a framework for conversion of data between geographical units. Int J Popul Geog
2002;8 69–82.
23. Norman P, Rees P, Boyle P. Achieving data compatibility over
space and time: creating consistent geographical zones. Int
JPopul Geog 2003;9:365–86.
24. Townsend P, Phillimore P, Beattie A. Health and Deprivation:
Inequality and the North. London: Croom Helm, 1988.
25. Norman P. Identifying change over time in small area socioeconomic deprivation. Appl Spatial Anal Policy 2010;3:107–38.
26. Toledano MB, Nieuwenhuijsen MJ, Best N et al. Relation of trihalomethane concentrations in public water supplies to stillbirth
and birth weight in three water regions in England. Environ
Health Perspect 2005;113:225–32.
27. The Stationery Office. Scottish Statutory Instruments, Water
Supply Water Quality (Scotland) Regulations No. 95. 2010.
http://www.legislation.gov.uk/ssi/2010/95/pdfs/ssi_20100095_
en.pdf (5 November 2013, date last accessed).
28. The Stationery Office. Statutory Instruments Water, England
and Wales, Water Supply Water Quality Regulations No. 991.
2010. http://dwi.defra.gov.uk/stakeholders/legislation/wsr2010.
pdf (5 November 2013, date last accessed).
29. Office for National Statistics. 2001 Census: Digitised Boundary
Data (England and Wales). Edinburgh: Census Geography Data
Unit, EDINA, University of Edinburgh, 2001.
30. General Register Office for Scotland. 2001 Census: Digitised
Boundary Data (Scotland). Edinburgh: Census Geography Data
Unit, EDINA, University of Edinburgh, 2001.
31. Environmental Systems Research Institute. ArcGIS Desktop:
Release 10. Redlands, CA: Environmental Systems Research
Institute, 2011.
32. Lee L. NADA: Nondetects and Data Analysis for Environmental
Data. R package version 1.5-3, 2010. http://cran.r-project.org/
src/contrib/Archive/NADA/ (5 November 2013, date last
accessed).
International Journal of Epidemiology, 2014, Vol. 43, No. 1
33. Core Team. R: A Language and Environment for Statistical
Computing. Vienna: R Foundation for Statistical Computing,
2013.
34. StataCorp. Stata Statistical Software: Release 12. College
Station, TX: StataCorp LP, 2011.
35. Akaike H. A new look at the statistical model identification.
IEEE Trans Automatic Control 1974;19:716–23.
36. Sterne JA, Davey Smith G. Sifting the evidence—what’s wrong
with significance tests? BMJ 2001;322:226–31.
37. Cohn PD. A Brief Report on the Association of Drinking Water
Fluoridation and the Incidence of Osteosarcoma Among Young
Males. Trenton, NJ: New Jersey Department of Health and
Environmental Health Services, 1992.
38. Comber H, Deady S, Montgomery E, Gavin A. Drinking water
fluoridation and osteosarcoma incidence on the island of Ireland.
Cancer Causes Control 2011;22:919–24.
39. Levy M, Leclerc BS. Fluoride in drinking water and osteosarcoma incidence rates in the continental United States among
children and adolescents. Cancer Epidemiol 2012;36:e83–e88.
40. National Toxicology Program (NTP). Toxicology and
Carcinogenesis Studies of Sodium Fluoride in F344/N Rats and
B6C3f1 Mice. Technical report Series No. 393. Research
Triangle Park, NC: National Institute of Environmental Health
Sciences, 1990.
41. Maurer JK, Cheng MC, Boysen BG, Anderson RL. Two-year
carcinogenicity study of sodium fluoride in rats. J Natl Cancer
Inst 1990;82:1118–26.
42. Hoover RN, Devesa SS, Cantor KP, Lubin JH, Fraumeni JF.
Time trends for bone and joint cancers and osteosarcomas in the
Surveillance, Epidemiology and End Results (SEER) Program.
In: National Cancer Institute. Benefits and Risks Report of the
Ad Hoc Committee on Fluoride of the Committee to Coordinate
Environmental Health and Related Programs. Washington, DC:
US Public Health Service, 1991.
43. Mahoney MC, Nasca PC, Bumett WS, Melius JM. Bone cancer
incidence rates in New York State: time trends and fluoridated
drinking water. Am J Public Health 1991;81:475–79.
44. Freni SC, Gaylor DW. International trends in the incidence of
bone cancer are not related to drinking water fluoridation.
Cancer 1992;70:611–18.
234
45. Moss ME, Kanarek MS, Anderson HA, Hanrahan LP, Remington PL. Osteosarcoma, seasonality and environmental factors in
Wisconsin, 1979-1989. Arch Environ Health 1995;50:235–41.
46. Gelberg KH, Fitzgerald EF, Hwang SA, Dubrow R. Fluoride exposure and childhood osteosarcoma: a case-control study. Am J
Public Health 1995;85:1678–83.
47. Bassin EB, Wypij D, Davis RB, Mittleman MA. Age-specific fluoride exposure in drinking water and osteosarcoma (United
States). Cancer Causes Control 2006;17:421–28.
48. Kim FM, Hayes C, Williams PL et al. National Osteosarcoma
Etiology Group. An assessment of bone fluoride and osteosarcoma. J Dent Res 2011;90:1171–76.
49. Ries LA, Smith MA, Gurney JG et al. (eds). Cancer Incidence
and Survival Among Children and Adolescents: United States
SEER Program 1975-1995, NIH Pub. No. 99-4649. Bethesda,
MD: National Cancer Institute, SEER Program, 1999.
50. Mirabello L, Troisi RJ, Savage SA. Osteosarcoma incidence and
survival rates from 1973 to 2004: data from the Surveillance,
Epidemiology,
and
End
Results
Program
Cancer
2009;115:1531–3.
51. Douglass CW, Joshipura K. Caution needed in fluoride and
osteosarcoma study. Cancer Causes Control 2006;17:481–82.
52. Sandhu R, Lal H, Kundu ZS, Kharb S. Serum fluoride and sialic
acid levels in osteosarcoma. Biol Trace Elem Res 2011;144:1–5.
53. Jackson PJ, Harvey PW, Young WF. Chemistry and Bioavailability Aspects of Fluoride in Drinking-Water. Report No. CO5037.
Marlow, Bucks, UK: Water Research Centre–National Sanitation Foundation (WRc-NSF) 2002.
54. Maguire A, Zohouri FV, Mathers JC, Steen IN, Hindmarch PN,
Moynihan PJ. Bioavailability of fluoride in drinking water: a
human experimental study. J Dent Res 2005;84:989–93.
55. Greenland S, Robins J. Invited commentary: ecological studies –
biases, misconceptions, and counterexamples. Am J Epidemiol
1994;139:747–60.
56. Wakefield J. Ecologic studies revisited. Annu Rev Public Health
2008;29:75–90.
57. Hac E, Czarnowski W, Gos T, Krechniak J. Lead and fluoride
content in human bone and hair in the Gdansk region. Sci Total
Environ 1997;206:249–54.